This paper aims to construct a multimodal poetry translation corpus for easy retrieval by poetry translation researchers and enthusiasts. In this paper, we combine the AdaBoost model and ELM network and propose the ELM-AdaBoost method to put the existing poetry translation corpus into the ELM network for learning, to obtain several weak predictors, and then use AdaBoost for classification iteration, and update the weights according to the prediction sequence weights, and then obtain strong predictors, and finally construct a multimodal poetry translation corpus. The search of this corpus shows that, in terms of the ideographic performance of the translations, the ancient poems perform the best, followed by the five-line stanzas, with mean evaluation scores of 83.2 and 80.9, respectively. The seven-line stanzas have the best phonetic performance, with an average rating of 73.2. The rhetoric of five-verse poems was the best, with an average rating of 63.5 marks. The overall translation effect is relatively poor because the meaning is often difficult to account for, or there is a cultural gap in the translation of poems. The multimodal translation corpus based on the AdaBoost model is a powerful tool for poetry translation research, which provides strong data support for Chinese poetry translation research and is of great significance for Chinese poetry culture.